On 10/29/06, Dan Price <dp at eng.sun.com> wrote:
> On Sun 29 Oct 2006 at 12:39PM, Mike Gerdts wrote:
> > 4967733 bad idea for zones to sync(1M) during shutdown
> > 6400646 zones induce between 0 and 6 calls to sync() on each reboot
> >
> > The attached patch addresses these bugs.  Contributor agreement is OS0018.
> I'll sponsor this one.   (Uhh, except that you forgot to attach the
> patch).

Good catch.  :)

> This is a great bug to fix.
> I think the remaining question we should investigate is whether
> sync(2) should do anything at all in a non-global zone (or should
> somehow be restricted in scope).  Probably we should check in with the
> VM and/or UFS folks about this.

Suppose that sync(2) became a no-op (or returned ENOSYS(?)) in a
non-global zone, does it make sense to keep at least the changes to
rc0.sh?  The only logic to keeping them would be to keep sysadmins
that read rc scripts from *thinking* that a server with 100 zones
shutting down is going to see 300 syncs.

The real motivation for this and the other fix I just sent to
request-sponsor was to figure out how to use mercurial.  As such, if
there is any goofiness with the patch or would you prefer "hg bundle"
over "hg export", please let me know.  One particular thing that I was
looking into was the impact of "zfs clone" instead of "hg clone" (no
impact seen so far...).  Any wisdom that can be shared on that front
would likewise be appreciated.


Mike Gerdts
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 4967733,6400646.patch
Type: application/octet-stream
Size: 1172 bytes
Desc: not available

Reply via email to